Augmented Semi-supervised Learning for CNN Based Automatic Modulation Classification

The classification of signal modulation using machine learning algorithms requires considerable amount of labeled signal. In this paper, a semi-supervised learning framework for a CNN based modulation classifier is proposed to reduce the amount of labeled signal needed while achieving good classification performance. Various techniques have been adopted to improve upon the classic semi-supervised learning framework for improved performance in modulation classification. The results show that the proposed method achieves accuracy close to the supervised classifier with large amount labeled signal while far superior to the one with the same amount of labeled signal.